AIДизайн8 мин. чтения

Создаем креативное AI-видео пошагово

Пошагово повторяем вирусное волшебное видео с собой в главной роли.

ИИ-видео набирают миллионы просмотров, а повторить их и сделать своё проще, чем ты думаешь. Давай покажу, как я сделала такое видео за один вечер.

Первое: всё начинается с вдохновения. Как-то мне попалось одно видео в ленте Инстаграма. Кроме того, что оно набрало миллион просмотров, оно вдохновило меня на создание своего видео, но в стиле мира Nexfield Academy. Потому что учиться у нас настоящее волшебство, зуб даю.

И я создала своё видео всего за один вечер и покажу тебе, как просто это сделать по шагам.

Готовое AI-видео: сова приносит письмо из Nexfield Academy, а впереди замок и фейерверк.

Шаг 1. Инструменты

Для создания видео нам понадобятся:

  • GPT Image для всех изображений: персонаж, кадры, конверт, сова.
  • Claude для написания промтов.
  • Seedance в Higgsfield для создания видео.

Шаг 2. Создаём лист персонажа

Первым делом нужен лист персонажа: ты крупно, в полный рост спереди и со спины. На нём держится всё остальное: именно по нему нейросеть будет узнавать твоё лицо в каждом кадре.

Исходное фото для AI-видео: девушка в чёрной футболке за столом с микрофоном и ноутбуком

Исходное фото, из которого собираем лист персонажа.

Берём своё фото, прикрепляем в GPT Image и вставляем этот промт:

Промт 1 · лист персонажа (GPT Image)

Используй промт

A professional character reference sheet for a video and AI-production workflow, built from the attached image. Image 1 is the identity anchor and the only source of truth for who this person is: reproduce the face, facial structure and proportions, skin tone and texture, eye colour, eyebrows, hair (length, colour, texture, parting and styling), facial hair, body build and any distinguishing features (freckles, moles, scars, tattoos, piercings, glasses, makeup) exactly as they appear in Image 1, identically in every panel. Take only the person's identity and appearance from Image 1 – not its pose, framing, camera angle, background, lighting, colour cast or filters. If Image 1 shows more than one person, use the most prominent one. Render in the same medium and level of realism as Image 1; for a photograph this means photorealistic, editorial studio quality. Background: one continuous seamless studio backdrop in clean neutral deep grey (#3a3a3c), no gradient banding, no props, no floor line. Layout: horizontal landscape composition in a 16:9 aspect ratio, read left to right, three vertical panels with even margins and consistent panel spacing. Panel 1 (leftmost, widest): chest-up portrait, front view, head and upper chest in frame, eyes looking straight into the camera, tack-sharp focus on the eyes with soft catchlights in both. Panel 2 (center): full-body front view, standing relaxed neutral A-pose, arms slightly away from the body, weight evenly distributed, head-to-toe fully inside the frame including the feet. Panel 3 (rightmost): full-body back view, same standing pose, showing the back of the hairstyle, back posture, garment fit and the backs of the shoes. Character: the same person from Image 1 in all three panels, with a calm, neutral, closed-mouth expression close to their resting expression in Image 1. Do not beautify, idealise, slim down, age or de-age the person, and do not alter any facial feature. Where Image 1 does not show something – the back of the head, the body, the legs, the feet – extend it logically and consistently from what is visible: the hairstyle continues naturally at the back, and height, build and body proportions match the face, neck, shoulders and any visible parts of the body. Skin texture stays true to Image 1; for a photograph, keep realistic natural skin with visible pores and subtle imperfections, no heavy retouching. Wardrobe: the exact outfit from Image 1 – same garments, colours, materials, fit and footwear – kept identical across all three panels. If the outfit is only partly visible, complete it in the same style, colours and materials. If no outfit is visible, dress the person in simple, plain, well-fitted neutral basics with minimal everyday shoes, no logos or prints. Lighting and render: soft even studio lighting, large diffused key light slightly to the front-left with gentle fill from the right, soft natural shadows under the chin and on the backdrop, no harsh speculars, neutral white balance, true-to-life skin tones matching Image 1, identical lighting and colour grading in all three panels. Full-frame camera with an 85mm lens look, controlled shallow depth of field, crisp detail, high dynamic range. Constraints: the same single person shown three times, no other people, no duplicated or averaged face – the identity must match Image 1 in all three panels. Do not restyle or recolour the hair. Do not add, remove or change facial hair, glasses, jewellery, piercings, tattoos or makeup – keep exactly what Image 1 shows. No text, no labels, no annotations, no arrows, no measurement lines, no watermark, no logos, no borders or frames around the panels, no cinematic colour grading or teal-orange tint. Aspect ratio 16:9

На выходе получаем лист персонажа в той одежде, что была на фото.

Лист персонажа из GPT Image: портрет, полный рост спереди и со спины на сером фоне

Лист персонажа: портрет, полный рост спереди и со спины.

Шаг 3. Переодеваем персонажа в брендовый аутфит

Теперь переодеваем себя в нужный аутфит, но в брендовых цветах Nexfield Academy. В GPT Image прикрепляем три картинки именно в таком порядке:

  1. Лист персонажа из шага 2.
  2. Референс аутфита: любое фото костюма, который тебе нравится.
  3. Референс цвета: брендовая палитра или логотип.

И вставляем промт:

Промт 2 · брендовый аутфит (GPT Image)

Используй промт

Edit Image 1, a three-panel character reference sheet. Keep everything about the person exactly as in Image 1: face, hair, skin, body, the pose in each panel, the layout, the grey backdrop, the lighting and the camera. Change only the wardrobe: dress the person in the outfit from Image 2, recolored into the brand palette from Image 3. Outfit: a knit headband with diagonal stripes, a crisp white collared shirt, a diagonally striped tie, a long chunky knit scarf with horizontal stripes and tasseled ends, a floor-length black hooded robe with colored lining on the hood and front edges, a small embroidered shield patch on the left chest, a short black pleated skirt, argyle knee-high socks with a solid cuff band, and glossy black chunky platform Mary Jane shoes. Colors: replace every accent color of the outfit in Image 2 with the palette from Image 3, royal blue and white. The stripes on the headband, tie and scarf become royal blue and white, the robe lining becomes bright royal blue, and the argyle socks become black and royal blue. The robe, skirt and shoes stay black, and the shirt stays white. Branding: the chest patch is a royal blue shield with a white serif letter N, and the same small N shield sits on the cuff band of each sock. Any other emblems, crests or lettering from Image 2 are replaced with this N shield. The outfit is identical across all three panels, front and back, with realistic fabric textures: chunky knit, crisp cotton, matte wool robe, glossy patent leather shoes. No text, labels or watermarks. Aspect ratio 16:9

Если у твоего бренда другие цвета, просто замени royal blue and white на свои, а букву N на свою монограмму.

Готово, персонаж переодет и выглядит так:

Лист персонажа в аутфите Nexfield Academy: чёрная мантия, сине-белые шарф, галстук и ободок

Тот же лист персонажа, но уже в аутфите в цветах Nexfield Academy.

Шаг 4. Первый кадр

Теперь нужен кадр, с которого начинается видео. В GPT Image прикрепляем лист персонажа в новом аутфите и референс локации и вставляем промт:

Промт 3 · первый кадр (GPT Image)

Используй промт

A cinematic vertical still of the woman from Image 1, keeping her exact face from the close-up portrait on the left of that sheet: same facial features, eyes, skin tone, any piercings, and the same long wavy hair. She wears exactly the outfit shown in Image 1, with every detail and color as on the sheet. Recreate the location from Image 2: a vast open green steppe at blue hour, lush emerald grass stretching to low rolling hills, a few white sheep grazing in the middle distance on the left, and a dramatic sky with towering slate-blue and periwinkle cumulus clouds and a thin band of peach-pink afterglow along the horizon. She stands full-length in the meadow, her body turned slightly to the left, calmly gazing toward the horizon on the left with a soft, dreamy expression, arms relaxed at her sides. A light evening breeze lifts a few strands of her hair and moves the scarf tassels and the hem of the robe, revealing a flash of the colored lining. Blue hour, about twenty minutes after sunset: soft, cool, even skylight on her face and clothing, with a subtle warm rim light from the horizon glow on her hair. The mood is serene and quietly magical, the calm moment before a fairy tale begins. Framed like a still from a vertical video shot on a phone, 35mm lens at chest height, natural photographic look with fine grain. She stands slightly right of center, fully visible from head to shoes, the horizon just below the middle of the frame, with generous open sky above her and in the upper right corner. Professional photography quality, ultra-high resolution, realistic skin, knit and fabric texture, technically perfect exposure. Aspect ratio 9:16

Пустое небо справа сверху оставляем специально: оттуда потом залетит сова.

Первый кадр AI-видео: девушка в мантии Nexfield Academy стоит в поле с овцами в синий час

Первый кадр: синий час, овцы на лугу и свободное небо справа для совы.

Шаг 5. Последний кадр

В GPT Image прикрепляем лист персонажа и референс замка, который тебе нравится.

Промт 4 · последний кадр с замком (GPT Image)

Используй промт

A wide cinematic vertical shot at late twilight. In the upper half of the frame, the castle from Image 2 crowns the top of a broad grassy hill: the same architecture, towers, spires and details as in Image 2, rising from a base of jagged grey rock with thin mist drifting around it, and dozens of windows glowing with warm golden light. It stands alone on the hilltop, surrounded only by grass, rock and mist. Above and behind the towers, a grand fireworks display bursts across the sky: large white-gold and pink chrysanthemum shells, glittering trails and slowly falling embers. In the middle ground, seen from behind and matching the back view in Image 1, the woman from Image 1 walks across the meadow toward the castle: her long wavy hair over a floor-length black hooded robe with colored lining in the hood, her headband, the tasseled ends of her striped scarf swinging at her sides, her argyle knee socks and platform shoes, every color exactly as on the sheet. Two white sheep graze on the grass to her left. In the lower-left foreground, a small winding stream curves through the grass, its rippling surface reflecting the glow of the fireworks and the lit castle windows. Lighting: late blue hour, a deep cobalt sky fading to soft lavender-pink clouds near the horizon, cool ambient light over the darkening meadow, warm golden light from the castle windows and sparkling light from the fireworks as accents. The mood is magical and triumphant, the moment a dream comes true. Composition: a phone camera at chest height a few meters behind her, the castle filling the upper half of the frame, fireworks across the top third, her figure at the center in the middle ground, the stream leading the eye from the bottom left into the scene. Photorealistic cinematic quality, ultra-high resolution, sharp architectural detail, technically flawless exposure. Aspect ratio 9:16

Последний кадр AI-видео: девушка бежит по лугу к замку на холме под фейерверком

Последний кадр: замок на холме, фейерверк и героиня на пути к нему.

Шаг 6. Конверт с печатью Nexfield Academy

Дальше детали, которых не было в референс-видео и которые я хотела добавить: конверт с печатью академии и сова. Начнём с конверта, референсы тут не нужны:

Промт 5 · конверт с печатью (GPT Image)

Используй промт

The back of a vintage kraft parchment envelope lying flat, seen from directly above, the whole envelope visible and centered on a pure white background. The paper is thick, slightly textured, warm beige with subtle fibers and a soft aged patina. The top flap has an elegant decorative edge with gentle wavy curves, folding down to a point in the center. On the flap, a heraldic crest is printed in dark sepia-brown ink in a vintage engraved line-art style with fine hatching. An arched ribbon banner above the shield reads "NEXFIELD ACADEMY" in classic serif capitals. The shield is divided into four quarters with an open book, a drafting compass, a fountain pen nib and a computer cursor arrow, all in the same engraved style. A small inner shield at the center holds a serif letter "N". A ribbon beneath the shield reads "DESIGN · AI · CREATIVE". At the point of the flap, a glossy royal blue wax seal holds the envelope closed: a thick, slightly irregular round blob of wax with soft melted edges, embossed with a serif letter "N" inside a shield outline, catching soft highlights. Soft, even studio light from the upper left, a gentle shadow under the flap edge, a soft glossy highlight on the wax. Product photography style, 100mm macro lens at f/8, everything sharp from edge to edge, paper and wax texture clearly visible. Professional photography quality, ultra-high resolution, crisp legible lettering. Aspect ratio 3:2

После генерации проверь надписи: нейросеть иногда путает буквы в мелком тексте. Белый фон нужен, чтобы в видео модель взяла только сам конверт.

Крафтовый конверт с гербом Nexfield Academy и синей сургучной печатью с буквой N

Конверт с гербом и печатью академии на белом фоне.

Шаг 7. Сова

Сова белая и огромная, как в Гарри Поттере. Референс не обязателен, промта хватит:

Промт 6 · белая сова (GPT Image)

Используй промт

A snowy white owl in full flight, captured mid-glide as a clean reference image. The owl is an adult snowy owl with dense, pure white plumage, a light scattering of small dark grey-brown speckles across its chest, back and the tops of its wings, a round flat face, a short black hooked beak half-hidden in soft feathers, and bright golden-yellow eyes looking straight into the camera. Its wings are spread wide in an elegant upswept V, each primary flight feather fanned out and clearly separated at the tips. The feathered legs and dark talons are tucked neatly under the body, and the short tail is slightly fanned. The owl glides toward the viewer at a three-quarter front angle, flying from right to left, body gently banking, as if it is about to sweep past the camera. The background is a smooth seamless studio backdrop in a soft desaturated slate-blue gradient, with the owl as the only subject so its silhouette reads perfectly. Lighting mimics the blue hour: soft, cool diffused light from above and the front evenly illuminates the white plumage, with gentle blue-grey shadows in the feather layers, plus a subtle warm peach rim light from behind tracing the upper edges of the wings. Wildlife photography style, 400mm f/2.8 lens at 1/4000s, the owl frozen tack-sharp, centered with its full wingspan inside the frame. Professional wildlife photography quality, ultra-high resolution, exceptional feather detail, technically perfect exposure. Aspect ratio 16:9

Фон делаем в цвет вечернего неба, чтобы сова по свету сразу вписалась в сцену.

Белая полярная сова в полёте с раскрытыми крыльями на сине-сером фоне

Сова в полёте: отдельный референс для видео.

Шаг 8. POV-кадр с письмом

И ещё одна деталь: я хотела добавить POV-ракурс на конверт. Он настолько волшебный, что хочется рассмотреть его поближе.

Прикрепляем лист персонажа и конверт. Лист нужен, чтобы сохранились все детали, которые попадут в кадр: рукава, шарф, галстук, гольфы и обувь.

Промт 7 · POV-кадр с письмом (GPT Image)

Используй промт

A first-person POV shot through the eyes of the woman from Image 1, looking straight down at a letter she has just caught. Both of her hands hold the envelope from Image 2 by its sides at waist height, with the crest and the wax seal facing up, the lettering crisp and legible, exactly as in Image 2. Her hands look natural and relaxed, with neat short nails and a skin tone matching the woman in Image 1, exactly two hands with five fingers each. She wears exactly the outfit from Image 1, with every color and detail as on the sheet: the wide black robe sleeves with colored lining and white shirt cuffs at the lower corners of the frame, the tasseled ends of the striped knit scarf hanging down on both sides, the tip of the striped tie at the bottom center. Below the envelope, on the grass, are her argyle knee socks with the shield on the cuffs and her glossy black platform Mary Jane shoes. She stands in lush green grass dotted with tiny white daisies. Soft, cool evening light falls evenly from above, with a gentle highlight on the wax seal. The mood is wonder and quiet anticipation, the moment a long-awaited letter finally arrives. Shot like a realistic handheld phone POV with a 24mm wide-angle lens, focus on the envelope, the envelope in the upper middle of the frame, the shoes below it. Professional photography quality, ultra-high resolution, realistic skin, paper, wax and fabric texture. Aspect ratio 9:16

POV-кадр: руки держат конверт Nexfield Academy с сургучной печатью над травой

POV-кадр: конверт в руках, внизу гольфы и туфли из аутфита.

Шаг 9. Режиссируем видео

Все детали готовы, теперь собираем их вместе. Описываем простыми словами все кадры по порядку. Описывай каждое действие максимально подробно: чем точнее описание, тем лучше ты управляешь видео. Вот моё описание:

Описание сцены для Claude

Используй промт

Персонаж стоит в кадре, всё начинается с первого кадра. Потом появляется сова, которая залетает в кадр справа с письмом и сбрасывает его. Человек ловит письмо и смотрит на него. Потом POV-кадр на письмо. Потом человек начинает бежать к замку, а сова продолжает лететь к замку. Над замком начинается фейерверк. Музыка волшебная, естественные звуки фейерверков и лёгкого ветра. Соотношение сторон 9:16.

Закидываем это описание в Claude и прикрепляем все изображения. Claude переведёт описание на понятный для Seedance язык, свяжет картинки через @ и скажет, в каком порядке их загружать:

ТегЧто загружаемКак отметить
@Image1Первый кадр (шаг 4)Start frame
@Image2Последний кадр (шаг 5)End frame
@Image3Лист персонажа в аутфите (шаг 3)Референс
@Image4Сова (шаг 7)Референс
@Image5Конверт (шаг 6)Референс
@Image6POV-кадр (шаг 8)Референс

И получаем вот такой промт:

Промт 8 · видео для Seedance, 15 с, 9:16

Используй промт

@Image1 as the first frame, @Image2 as the last frame. @Image3 is the woman: keep her face, hair and outfit exactly as in @Image3 throughout the video. @Image4 is the snowy white owl: keep its white plumage, dark speckles and golden eyes exactly as in @Image4. @Image5 is the letter: a kraft parchment envelope with the NEXFIELD ACADEMY crest and a royal blue wax seal, keep its look exactly as in @Image5. @Image6 is the composition of the POV shot. Vertical 9:16, 15 seconds, three shots joined by hard cuts. Realistic footage filmed on a phone held in someone's hand: the frame gently bobs and sways with the operator's breathing and steps, small natural shakes, quick reframing, like real amateur travel footage. The camera is never locked off. SHOT 1 [0-3.5s], handheld, full-length: [0-1s] Blue hour on a vast green steppe. She stands full-length in the meadow, gazing toward the horizon on the left. A light breeze moves her hair, the scarf tassels and the hem of her black robe. Behind her, the sheep slowly walk and graze. The handheld frame drifts slightly. [1-2.5s] The owl from @Image4 flies into the frame from the right, close to the camera, carrying the envelope from @Image5 in its talons, its wide white wings beating slowly, feathers crisp, wingtips slightly motion-blurred. She looks up at it. The owl glides just above her head and releases the envelope. [2.5-3.5s] The envelope flutters down, and she raises both hands and catches it in front of her chest. She holds it with both hands by its edges and looks down at it with a surprised, delighted smile. The owl keeps flying to the left toward the hills. HARD CUT. SHOT 2 [3.5-5.5s], first-person POV through her eyes, framed exactly as in @Image6: Looking down at her own hands holding the envelope with both hands, her sleeves, scarf ends, socks and shoes visible as in @Image6. A slow, gentle zoom in toward the crest and the royal blue wax seal. The POV moves with soft breathing and tiny natural hand tremors. HARD CUT. SHOT 3 [5.5-15s], handheld, continuous: [5.5-7s] Back to the side view: she looks up toward the owl flying away, presses the letter to her chest and starts walking to the left, then breaks into a run, robe and striped scarf streaming behind her. The camera pans left to follow her. [7-8.5s] The camera swings around until it is behind her, tracking her as she runs across the meadow toward the hills. Far ahead, the owl keeps flying. [8.5-10s] The camera pans slowly left and tilts up slightly: the castle from @Image2 slides into frame from the left edge and settles at the center on its rocky hill, mist around the base, warm golden light in the windows. The owl, now a small white shape, flies toward the castle towers and disappears among them. A small winding stream enters the lower-left foreground, two sheep graze on the slope, and the sky deepens into violet-blue with pink-lit clouds. The first rockets streak up from behind the castle on thin glowing trails. [10-15s] The rockets burst into large white-gold and pink chrysanthemum shells above the towers, and the fireworks grow into a full, continuous show that keeps going right up to the final frame: new rockets every second, overlapping bursts, fountains of golden sparks behind the towers, glittering embers drifting down, their light reflecting in the stream. She slows to a walk and continues toward the castle while the sheep slowly graze nearby. The camera settles into a wide shot behind her with a gentle handheld sway and lands exactly on @Image2. Audio: a soft magical orchestral score with gentle strings, celesta and chimes, swelling warmly as the fireworks begin. Natural sounds underneath: light evening wind and rustling grass, a soft whoosh of the owl's wings, a paper flutter as she catches the letter, and natural fireworks pops and crackles that continue until the very end.

Шаг 10. Генерируем в Higgsfield: сначала драфт

Идём в Higgsfield: Видео, модель Seedance, режим Draft. Это важно, всегда делай так. Драфт позволяет дёшево сгенерировать видео и посмотреть на концепт.

Higgsfield: Seedance 2.5 Draft, 15 секунд, 9:16, 480p, генерация драфта за 45 кредитов

Режим Draft: ролик на 15 секунд в 480p за 45 кредитов.

  1. Загружаем картинки в порядке из таблицы выше, вставляем промт и запускаем драфт.
  2. Смотрим результат. Если нужны правки, возвращаемся в Claude и описываем, что не так: например, сова летит не в ту сторону или конверт как-то не так падает. У меня было и то, и другое. Claude перепишет промт, и мы снова запускаем драфт.
  3. Когда драфт нравится, перегенерируем именно это видео в 1080p. Оно получится точно таким же, ничего не поплывёт.

Мне это сильно сэкономило кредитов. Драфт на 15 секунд стоил 45 кредитов, а обычная генерация сразу в хорошем качестве 240. Разница ощутимая, особенно когда переделываешь несколько раз. Больше приёмов собрали в статье о том, как экономить кредиты на генерации видео.

Результат

Готово, вот такой результат получился:

Финальное видео в вертикальном формате 9:16.

Используй его так, как твоей фантазии угодно. Нужно видео для сайта? Поменяй формат на горизонтальный 16:9. Для соцсетей оставь 9:16, как у меня, и публикуй как вертикальный контент.

А если хочешь научиться генерировать видео, освоить мастерство и самые нужные инструменты понятным языком, приходи в Nexfield Academy на курс «Видеопродакшн в AI». Там мы собираем полноценные видео в нейросетях: от сценария и раскадровки до камеры, звука и графики.

Подписывайся на наш регулярный дайджест

Будем присылать полезные статьи и новости из мира AI, дизайна, технологий и карьерного развитиядизайна и технологий

Более 3000 человек уже получают полезные материалы